Evaluation of CyberKnife® Fiducial Tracking Limitations to Assist Targeting Accuracy: A Phantom Study with Fiducial Displacement.

Introduction  The underlying assumptions of the CyberKnife® (Accuray, Sunnyvale, CA, US) fiducial tracking system are: i) fiducial positions are accurately detected; ii) inter-fiducial geometry remains consistent (rigid); iii) inter-fiducial geometric array changes are detected and either accommodated with corrections or treatment is interrupted. However: i) soft-tissue targets are deformable & fiducial migration is possible; ii) the accuracy of the tracking system has not previously been examined with fiducial displacement; iii) treatment interruptions may occur due to inter-fiducial geometric changes, but there is little information available to assist subsequent troubleshooting. The purpose of this study was to emulate a clinical target defined with a two, three, or four-fiducial array where one fiducial is displaced to mimic a target deformation or fiducial migration scenario. The objectives: evaluate the fiducial positioning accuracy, array interpretation, & corresponding corrections of the CyberKnife system, with the aim of assisting troubleshooting following fiducial displacement. Methods A novel solid-water phantom was constructed with three fixed fiducials (F1,F2,F3) & one moveable fiducial (F4), arranged as if placed to track an imaginary clinical target. Using either two fiducials (F1,F4), different combinations of three fiducials (F1,F2,F4; F1,F3,F4; F2,F3,F4) or four fiducials (F1,F2,F3,F4), repeat experiments were conducted where F4 was displaced inferiorly at 2-mm intervals from 0-16 mm. Data were acquired at each position of F4, including rigid body errors (RBE), fiducial x, y, & z coordinate displacements, six degrees of freedom (DOF) corrections, & robot center-of-mass (COM) translation corrections. Results Maximum positioning difference (mean±SD) between the reference and live x, y, & z coordinates for the three fixed fiducials was 0.08±0.30 mm, confirming good accuracy for fixed fiducial registration. For two fiducials (F1,F4), F4 registration was accurate to 14-mm displacement and the F4 x-axis coordinate change was 2.0±0.12 mm with each 2 mm inferior displacement validating the phantom for tracking evaluation. RBE was >5 mm (system threshold) at 6-14 mm F4 displacement: however, F1 was misidentified as the RBE main contributor. Further, F1/F4 false-lock occurred at 16 mm F4 displacement with corresponding RBE <3 mm & COM corrections >13 mm. For combinations of three fiducials, F4 registration was accurate to 10-mm displacement. RBE was >5 mm at 6-16 mm F4 displacement: however, F4 false-lock occurred at 12-16 mm with RBE 5-6 mm. For four fiducials, F4 registration was accurate to 4 mm displacement: however, F4 false-lock occurred at 6-16 mm displacement with concerning RBE <2 & <5 at 6 & 8-mm F4 displacement, respectively. False-locks were easily identified in the phantom but frequently uncorrectable. Conclusions Results indicate fiducial positioning accuracy and system output following fiducial displacement depends on the number of fiducials correlated, displacement distance, and clinical thresholds applied. Displacements ≤4 mm were accurately located, but some displacements 6-16 mm were misrepresented, either by erroneous main contributor (two-fiducial array only) or by false-locks and misleading RBE, which underestimated displacement. Operator vigilance and implementation of our practical guidelines based on the study findings may help reduce targeting error and assist troubleshooting in clinical situations.